• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.706-708
• /
• 2023

In this study, we embark on a journey to uncover the essence of emotions by exploring the depths of transfer learning on three pre-trained transformer models. Our quest to classify five emotions culminates in discovering the KLUE (Korean Language Understanding Evaluation)-BERT (Bidirectional Encoder Representations from Transformers) model, which is the most exceptional among its peers. Our analysis of F1 scores attests to its superior learning and generalization abilities on the experimental data. To delve deeper into the mystery behind its success, we employ the powerful SHAP (Shapley Additive Explanations) method to unravel the intricacies of the KLUE-BERT model. The findings of our investigation are presented with a mesmerizing text plot visualization, which serves as a window into the model's soul. This approach enables us to grasp the impact of individual tokens on emotion classification and provides irrefutable, visually appealing evidence to support the predictions of the KLUE-BERT model.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.461-464
• /
• 2023

최근 GPT-3 와 LLaMa 같은 생성형 거대 언어모델을 활용한 서비스가 공개되었고, 실제로 많은 사람들이 사용하고 있다. 해당 모델들은 사용자들의 다양한 질문에 대해 유창한 답변을 한다는 이유로 주목받고 있다. 하지만 LLMs 의 답변에는 종종 Inconsistent content 와 non-factual statement 가 존재하며, 이는 사용자들로 하여금 잘못된 정보의 전파 등의 문제를 야기할 수 있다. 이에 논문에서는 동일한 질문에 대한 LLM 의 답변 샘플과 외부 지식을 활용한 Hallucination Detection 방법을 제안한다. 제안한 방법은 동일한 질문에 대한 LLM 의 답변들을 이용해 일관성 점수(Consistency score)를 계산한다. 거기에 외부 지식을 이용한 사실검증을 통해 사실성 점수(Factuality score)를 계산한다. 계산된 일관성 점수와 사실성 점수를 활용하여 문장 수준의 Hallucination Detection 을 가능하게 했다. 실험에는 GPT-3 를 이용하여 WikiBio dataset 에 있는 인물에 대한 passage 를 생성한 데이터셋을 사용하였으며, 우리는 해당 방법을 통해 문장 수준에서의 Hallucination Detection 성능이 baseline 보다 AUC-PR scores 에서 향상됨을 보였다.

• Transactions on Semiconductor Engineering
• /
• v.1 no.1
• /
• pp.23-30
• /
• 2023

With the recent development of Generative AI technology by IT giants, the size of the transformer model is increasing exponentially over trillion won. In order to continuously enable these AI services, it is essential to reduce the weight of the model. In this paper, we find a hardware-friendly structured pruning pattern and propose a lightweight method of the transformer model. Since compression proceeds by utilizing the characteristics of the model algorithm, the size of the model can be reduced and performance can be maintained as much as possible. Experiments show that the structured pruning proposed when pruning GPT-2 and BERT language models shows almost similar performance to fine-grained pruning even in highly sparse regions. This approach reduces model parameters by 80% and allows hardware acceleration in structured form with 0.003% accuracy loss compared to fine-tuned pruning.

• International Journal of Advanced Culture Technology
• /
• v.11 no.4
• /
• pp.385-392
• /
• 2023

The smart tourism service provides tourists with personal travel planner services and context-awareness-based tour guide services. In this paper, we propose the personal travel planner service that creates my travel itinerary using the smart tourism app and the travel planner system. The smart tourism app provides recommended travel products and POI tourist information used to create my travel itinerary. The smart tourism app also provides the smart tourism chatbot service that allows users to select POI tourist information easily and conveniently. The travel planner system consists of the smart tourism information system and the smart tourism chatbot system. The smart tourism information system provides users with travel planner services, recommended travel products, and POI tourism information through the smart tourism app. The smart tourism chatbot system consists of named entity recognition (NER), dialogue state tracking (DST), and Neo4J servers, and provides chatbot services as a smart tourism app. Users can create their own travel itinerary, modify the travel itinerary while traveling, and then register it as a recommended travel product to users, including acquaintances.

• Journal of Information Technology Applications and Management
• /
• v.31 no.3
• /
• pp.149-163
• /
• 2024

As conversational AI systems such as ChatGPT have become more advanced, researchers are exploring ways to use them in education. However, we need effective ways to evaluate these systems before allowing them to help teach students. This study proposes a detailed framework for testing conversational AI across three important criteria as follow. First, specialized benchmarks that measure skills include giving clear explanations, adapting to context during long dialogues, and maintaining a consistent teaching personality. Second, adaptive standards check whether the systems meet the ethical requirements of privacy, fairness, and transparency. These standards are regularly updated to match societal expectations. Lastly, evaluations were conducted from three perspectives: technical accuracy on test datasets, performance during simulations with groups of virtual students, and feedback from real students and teachers using the system. This framework provides a robust methodology for identifying strengths and weaknesses of conversational AI before its deployment in schools. It emphasizes assessments tailored to the critical qualities of dialogic intelligence, user-centric metrics capturing real-world impact, and ethical alignment through participatory design. Responsible innovation by AI assistants requires evidence that they can enhance accessible, engaging, and personalized education without disrupting teaching effectiveness or student agency.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2007.11a
• /
• pp.1226-1229
• /
• 2007

XML(eXtensible Markup Language)은 문서구조나 내용, 스타일의 자유로운 표현, 웹상에서 구조화된 문서의 전송이 가능하도록 설계된 표준 마크업 언어로, HTML에 비해 인터넷상에서의 정보 및 문서교환, 정보 검색 둥 이 편리한 장점을 갖는다. 그러나 XML 데이터가 방대해지고 복잡해짐에 따라 접근제어 정책을 규정하고 수행하기 위한 모델과 메커니즘의 필요에 따라 RBAC를 기반으로 하는 모델과 메커니즘에 대한 연구가 이루어지고 있다. 그러나 기존 연구들은 개념적 모델들 위주로 연구의 초점이 되어 있어 실제 모델 구현 시 시간적 공간적 오버헤드가 발생하는 단점이 있다. 따라서 본 연구에서는 XML 문서의 RBAC 기법을 기반으로 하는 노드의 권한 정보를 가지고 있는 접근 권한 트리 모델을 제안하였다. 이를 이용하여 노드권한에 모든 정보를 저장하고 권한을 가지는 구성요소에 접근할 경우나 권한을 추가로 부여하는 경우에 XML 엘리먼트, 속성, 링크 등의 세부요소에 대한 권한 체크 시간을 감소시켜 사용자에게 빠른 뷰를 제공할 수 있는 성능이 우수한 모델을 제시 하고자 한다.

• ETRI Journal
• /
• v.46 no.1
• /
• pp.127-136
• /
• 2024

The Korean language has written (formal) and spoken (phonetic) forms that differ in their application, which can lead to confusion, especially when dealing with numbers and embedded Western words and phrases. This fact makes it difficult to automate Korean speech recognition models due to the need for a complete transcription training dataset. Because such datasets are frequently constructed using broadcast audio and their accompanying transcriptions, they do not follow a discrete rule-based matching pattern. Furthermore, these mismatches are exacerbated over time due to changing tacit policies. To mitigate this problem, we introduce a data-driven Korean spoken-to-written transcription conversion technique that enhances the automatic conversion of numbers and Western phrases to improve automatic translation model performance.

• Research in Plant Disease
• /
• v.30 no.1
• /
• pp.99-102
• /
• 2024

Integrated pest management is essential for controlling plant diseases that reduce crop yields. Rapid diagnosis is crucial for effective management in the event of an outbreak to identify the cause and minimize damage. Diagnosis methods range from indirect visual observation, which can be subjective and inaccurate, to machine learning and deep learning predictions that may suffer from biased data. Direct molecular-based methods, while accurate, are complex and time-consuming. However, the development of large multimodal models, like GPT-4, combines image recognition with natural language processing for more accurate diagnostic information. This study introduces GPT-4-based system for diagnosing plant diseases utilizing a detailed knowledge base with 1,420 host plants, 2,462 pathogens, and 37,467 pesticide instances from the official plant disease and pesticide registries of Korea. The AI plant doctor offers interactive advice on diagnosis, control methods, and pesticide use for diseases in Korea and is accessible at https://pdoc.scnu.ac.kr/.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2024.05a
• /
• pp.490-493
• /
• 2024

현재 다양한 분야에서 AI 가 사용되고 있다. 최근에는 소프트웨어공학 관점에서 요구 사항 분석에 Chat GPT 와 같은 LLM 모델을 적용하고 있다. 하지만 1) 대부분의 생성형 AI 는 불투명한 공정을 통해 3D 이미지가 생성하고, 3D 이미지를 생성할 때마다 다른 이미지를 생성한다. 이에 따라 동일한 인물이나 사물을 사용하고 싶은 사용자들은 동일한 객체가 들어간 그림을 일관성 있게 생성할 수 없다. 2) 또한 LLM 과 이미지 생성 AI 와의 결합이 시도 되고 있지만 문장 의미 분석 성능이 부족하다. 이를 해결하기 위해, 자연어 요구사항을 언어학적 기법을 통해 분석하고, 분석 결과를 기반으로 UML 시퀀스 다이어그램 및 3D 객체 생성 메커니즘을 제안한다. 즉 언어학적 분석 기법을 통해, 요구사항의 정확한 의미와 속성을 추출한다. 그런 다음 추출된 정보를 시퀀스 다이어그램과 매핑하여 3D 객체 이미지를 생성한다. 제안하는 방법을 통해 3D 객체 생성의 소프트웨어 개발 공정 사용으로 생산성을 높여 시간과 비용을 단축할 수 있을 것으로 기대한다.

• Phonetics and Speech Sciences
• /
• v.16 no.2
• /
• pp.1-9
• /
• 2024

This study investigates the fixed and random factors affecting response patterns of wh-scope interpretations in Gyeongsang Korean. It employed logistic mixed-effects regression models to analyze responses from 24 participants who listened to 40 pre-recorded stimuli from 40 different speakers. The stimuli consisted of an embedded wh-phrase and an interrogative ending marker, "-nkiko," thereby forming a wh-question, specifically a matrix wh-scope. Participants repeated the test three times. The study found that the prominence level of a prosodic phrase composed of an embedded verb and a complementizer was inversely related to responses with wh-questions, as demonstrated through multiple regression analysis in Yun. The test trial significantly impacted the number of responses with wh-questions, increasing from 50.3% in the first trial to 58.8% and 61.2% in subsequent trials. Examination of random subject effects revealed two main factors influencing responses: morpho-syntactic constraints and prosodic structural integrity. These two factors demonstrated the potential to be inversely weighted. Analysis of random stimulus effects suggested that the prominence level had limited effects on response patterns with each stimulus primarily eliciting one type of responses across trials.